CoChat

Maximizing the Information Content of Experiments in Systems Biology

Open access

The DOI matches its Crossref record; 63 of the 63 references it cites were found in a registry or an index; 4 of the 24 sources its claims cite were checked for support.

Abstract

Our understanding of most biological systems is in its infancy. Learning their structure and intricacies is fraught with challenges, and often side-stepped in favour of studying the function of different gene products in isolation from their physiological context. Constructing and inferring global mathematical models from experimental data is, however, central to systems biology. Different experimental setups provide different insights into such systems. Here we show how we can combine concepts from Bayesian inference and information theory in order to identify experiments that maximize the information content of the resulting data. This approach allows us to incorporate preliminary information; it is global and not constrained to some local neighbourhood in parameter space and it readily yields information on parameter robustness and confidence. Here we develop the theoretical framework and apply it to a range of exemplary problems that highlight how we can improve experimental investigations into the structure and dynamics of biological systems and their behavior.

Overview

CoChat Summary

TL;DR

The paper develops a Bayesian experimental design framework that uses mutual information to select experiments maximizing information about model parameters or predictions in systems biology. It applies the method to repressilator, Hes1, and EGF-dependent AKT models, showing that selected experiments improve parameter inference or predictive certainty.

Key findings

  • In the repressilator model, two of the four proposed experiments had the highest mutual information; ABC inference confirmed that data from those experiments allowed all parameters to be estimated, whereas data from the original repressilator system allowed only some parameters to be inferred with confidence.
  • In the Hes1 model, mRNA measurements carried more information about all parameters than total cellular Hes protein measurements; entropy-difference analysis on simulated data was consistent, although one protein-parameter result diverged for a specific dataset because mutual information is an average over possible data.
  • In the EGF-dependent AKT pathway, predictions under a noisy 5-pulse EGF input were highly uncertain using only the initial dataset; mutual information selected a step stimulus as most informative, and adding its outcome improved predictions and inferred EGFR turnover and EGFR initial condition at the required precision.

Methods

  • The framework combines Bayesian inference and information theory: it selects experiments by maximizing mutual information between parameters and experimental output, or between a target prediction and candidate experiment output, with Monte Carlo estimation.
  • Posterior parameter distributions are inferred using Approximate Bayesian Computation with sequential Monte Carlo (ABC-SMC), implemented in ABC-SysBio.
  • Models include a repressilator ODE toy model with hypothetical experiments, a three-component Hes1 ODE model, and an EGF-dependent AKT pathway model using experimentally measured normalized Western blot data from the Kuroda lab; computations used GPU-parallel ODE solving and Monte Carlo estimation.

Limitations

  • Stated: the approach chooses an optimal experiment from a finite, discrete set; continuous experimental design requires a different approach. Entropy estimates are histogram-based and biased, though comparisons across experiments remain valid if bin number and sample size are kept the same. Mutual information is an average over possible data, so it can diverge from entropy reduction for a specific dataset, as observed in the Hes1 example.
  • Evident from the text: the repressilator experiments are hypothetical, and the AKT data are normalized Western blot measurements; the paper also notes that Bayesian posterior computation via MCMC/SMC is costly.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1002888

Cites

63 of 63 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Information content of experimental data and flow chart of the experimental design method. (A) The regions of plausible parameters values for three different experiments. Each ellipse defines the set of parameters which are commensurate with the output of an experiment . In this example, the data leads to the most precise inference of the parameters. The parameters which explain the output of all the three experiments are at the intersection of the three ellipsoids. (B) Flowchart of the experimental design method. Given a mathematical model of the biological system, a set of experiments and the target information —which can be either a set of parameters to infer or a description of the experiment to predict — the Bayesian Experimental Design method determine the experiment to carry out. Once the experiment has been performed, the experimental data are then used to provide target information and to improve the model. Thereafter, the process can be iterated to select other experiments in order to improve the accuracy of the target information. (C) Link between the total and conditional entropies and the mutual information of experimental data and parameters .
Figure 1 Information content of experimental data and flow chart of the experimental design method. (A) The regions of plausible parameters values for three different experiments. Each ellipse defines the set of parameters which are commensurate with the output of an experiment . In this example, the data leads to the most precise inference of the parameters. The parameters which explain the output of all the three experiments are at the intersection of the three ellipsoids. (B) Flowchart of the experimental design method. Given a mathematical model of the biological system, a set of experiments and the target information —which can be either a set of parameters to infer or a description of the experiment to predict — the Bayesian Experimental Design method determine the experiment to carry out. Once the experiment has been performed, the experimental data are then used to provide target information and to improve the model. Thereafter, the process can be iterated to select other experiments in order to improve the accuracy of the target information. (C) Link between the total and conditional entropies and the mutual information of experimental data and parameters .CC BY · PubMed Central
Figure 2 The repressilator model (as described in [40]) and the set of possible experiments. (A) Illustration of the original repressilator model. The model consists of 3 mRNA species (coloured wavy lines, labeled , and ) and their corresponding proteins (circles shown in the same color with labels , and ). The 3 regulatory DNA regions are not modeled explicitly but included in the mRNA production process. They are shown for illustration purpose only. (B) The ordinary differential equations which describe the evolution of the concentration of the mRNAs and proteins over time. (C) Four potential modifications of the wild-type model. For each experimental intervention the modified parameters are listed (colours are as in A). The modifications of the wild-type model consist of decreasing one or several of the parameters of the system: in sets , and , the regime of the parameter is changed; in sets , and , respectively, parameters , and are modified for only one gene which breaks the symmetry of the system.
Figure 2 The repressilator model (as described in [40]) and the set of possible experiments. (A) Illustration of the original repressilator model. The model consists of 3 mRNA species (coloured wavy lines, labeled , and ) and their corresponding proteins (circles shown in the same color with labels , and ). The 3 regulatory DNA regions are not modeled explicitly but included in the mRNA production process. They are shown for illustration purpose only. (B) The ordinary differential equations which describe the evolution of the concentration of the mRNAs and proteins over time. (C) Four potential modifications of the wild-type model. For each experimental intervention the modified parameters are listed (colours are as in A). The modifications of the wild-type model consist of decreasing one or several of the parameters of the system: in sets , and , the regime of the parameter is changed; in sets , and , respectively, parameters , and are modified for only one gene which breaks the symmetry of the system.CC BY · PubMed Central
Figure 3 Experiment choice for parameter inference in the repressilator model. Top: The mutual information between the parameters and the output of each set of experiment (in dark green), and the entropy difference between the prior distribution and the posterior distribution. The posterior distribution is based on data obtained from simulation of the system for each experiment (in light green). The error bars on the mutual information barplots show the variance of the mutual information estimations over independent simulations. Bottom: For each set of experiment we show the histogram of the marginal of the posterior distribution of every parameters. The red line indicates the true parameter value.
Figure 3 Experiment choice for parameter inference in the repressilator model. Top: The mutual information between the parameters and the output of each set of experiment (in dark green), and the entropy difference between the prior distribution and the posterior distribution. The posterior distribution is based on data obtained from simulation of the system for each experiment (in light green). The error bars on the mutual information barplots show the variance of the mutual information estimations over independent simulations. Bottom: For each set of experiment we show the histogram of the marginal of the posterior distribution of every parameters. The red line indicates the true parameter value.CC BY · PubMed Central
Figure 4 Experiment selection for parameter inference in the Hes model. (A) Diagram of the Hes model and the associated ordinary differential equations. The model shows a cell (green box) with its nucleus (white circle). Hes mRNA (wavy line with label ) is produced inside the nucleus and translated into Hes cellular protein (), which in turn is then transported into the nucleus and becomes nuclear Hes protein (). is regulating the production of . (B) The mutual information between each parameters and the output of the system for respectively mRNA measurement (left) and total cellular Hes measurement (right). The barplot represents the mean and the variance over repetitions of the Monte-Carlo estimation. (C) Estimation of the difference between the entropy of the parameter prior and the entropy of the posterior distribution given one dataset. The dataset results from simulation of the Hes system. The parameter was set to be 0.03 min−1 as experimentally determined by [42].
Figure 4 Experiment selection for parameter inference in the Hes model. (A) Diagram of the Hes model and the associated ordinary differential equations. The model shows a cell (green box) with its nucleus (white circle). Hes mRNA (wavy line with label ) is produced inside the nucleus and translated into Hes cellular protein (), which in turn is then transported into the nucleus and becomes nuclear Hes protein (). is regulating the production of . (B) The mutual information between each parameters and the output of the system for respectively mRNA measurement (left) and total cellular Hes measurement (right). The barplot represents the mean and the variance over repetitions of the Monte-Carlo estimation. (C) Estimation of the difference between the entropy of the parameter prior and the entropy of the posterior distribution given one dataset. The dataset results from simulation of the Hes system. The parameter was set to be 0.03 min−1 as experimentally determined by [42].CC BY · PubMed Central
Figure 5 The EGF-dependent AKT pathway and an initial dataset. (A) Diagram of the model of the EGF-dependent AKT pathway. Epidermal growth factor (EGF, red triangle) is a stimulus for a signalling cascade, which results in the phosphorylation (green circle) of Akt (blue square) and S6 (purple square). EGF binds to the EGF membrane receptor EGFR (orange), which is generated from a pro-EGFR. The Binding results in the phosphorylation of the receptor, which consequently leads to the activation of downstream cascades (thick black circle). This simplified model was shown to capture the experimentally determined dynamics [45]. (B) A impulse input of EGF over seconds with an intensity of ng/ml (top) and the resulting time course of phosphorylated EGF receptor (pEGFR), phosphorylated Akt (pAKT) and phosphorylated S6 (pS6) in response to this stimulus (bottom). Data were provided by the authors of [45].
Figure 5 The EGF-dependent AKT pathway and an initial dataset. (A) Diagram of the model of the EGF-dependent AKT pathway. Epidermal growth factor (EGF, red triangle) is a stimulus for a signalling cascade, which results in the phosphorylation (green circle) of Akt (blue square) and S6 (purple square). EGF binds to the EGF membrane receptor EGFR (orange), which is generated from a pro-EGFR. The Binding results in the phosphorylation of the receptor, which consequently leads to the activation of downstream cascades (thick black circle). This simplified model was shown to capture the experimentally determined dynamics [45]. (B) A impulse input of EGF over seconds with an intensity of ng/ml (top) and the resulting time course of phosphorylated EGF receptor (pEGFR), phosphorylated Akt (pAKT) and phosphorylated S6 (pS6) in response to this stimulus (bottom). Data were provided by the authors of [45].CC BY · PubMed Central
Figure 6 Experiment selection for prediction in the EGF-dependent AKT pathway. (A) The noisy -impulses EGF input signal: the pulses are of intensity ng/ml and length seconds spaced by seconds with an additive background noise which is the absolute value of a gaussian white noise of variance . (B) The predicted time course of the proteins pEGFR, pAKT and pS6 under the noisy -impulses EGF input signal based on the initial dataset. (C) The mutual information between the time course of the species of interest under the noisy -impulses EGF input signal and the time course of the species under each of the following possible experiments: an impulse stimulus of length seconds with possible intensity (, , , and ng/ml), a step stimulus of length minutes with possible intensity (, , and ng/ml) and a ramp stimulus of length minutes with possible final intensity (, and ng/ml). (D) The predicted time course of the proteins pEGFR, pAKT and pS6 under the noisy -impulses EGF input signal based on the outcome of the step stimulus with intensity ng/ml, which is the experiment with the highest mutual information. The scale of the y-axis is different for Figures (B) and (D). (E) The posterior distribution of two parameters (EGFR turnover and EGFR initial condition) when using the initial dataset alone (left) and when using the initial dataset and the outcome of the step stimulus with intensity ng/ml (right). The scale of the EGFR turnover is the prior range for the figure on the left panel whereas it is times smaller for the figure in the right panel.
Figure 6 Experiment selection for prediction in the EGF-dependent AKT pathway. (A) The noisy -impulses EGF input signal: the pulses are of intensity ng/ml and length seconds spaced by seconds with an additive background noise which is the absolute value of a gaussian white noise of variance . (B) The predicted time course of the proteins pEGFR, pAKT and pS6 under the noisy -impulses EGF input signal based on the initial dataset. (C) The mutual information between the time course of the species of interest under the noisy -impulses EGF input signal and the time course of the species under each of the following possible experiments: an impulse stimulus of length seconds with possible intensity (, , , and ng/ml), a step stimulus of length minutes with possible intensity (, , and ng/ml) and a ramp stimulus of length minutes with possible final intensity (, and ng/ml). (D) The predicted time course of the proteins pEGFR, pAKT and pS6 under the noisy -impulses EGF input signal based on the outcome of the step stimulus with intensity ng/ml, which is the experiment with the highest mutual information. The scale of the y-axis is different for Figures (B) and (D). (E) The posterior distribution of two parameters (EGFR turnover and EGFR initial condition) when using the initial dataset alone (left) and when using the initial dataset and the outcome of the step stimulus with intensity ng/ml (right). The scale of the EGFR turnover is the prior range for the figure on the left panel whereas it is times smaller for the figure in the right panel.CC BY · PubMed Central

Verification

Claims the paper makes

3 Valid · 0 Not valid · 15 Not checked · 53 Own results and interpretations

41 more claims

Create a free account to see all 71 claims and to verify the 20 references still open.

See all claims — free account
  • Probably not a substantial claim. 14
  • It may state two separate claims as one. 15

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 63 references

Research works printed but not linked: 1

  1. J Bazil, 2012, A global parallel model based design of experiments method to minimize model output uncertainty, Bulletin of mathematical biology, 74, 1, 10.1007/s11538-011-9686-9 · DOI · Paper page

Cited by

See all 170

Metadata

Authors
Liepe, J., Filippi, S., Komorowski, M., & Stumpf, M. P. H.
Affiliations
  • Imperial College London (GB)
  • Institute of Fundamental Technological Research (PL)
Corresponding author
  • Liepe, Juliane
  • Filippi, Sarah
  • Komorowski, Michał
  • Stumpf, Michael P. H.
Venue
PLoS Computational Biology
Published
2013-01-31
Volume
9
Issue
1
Pages
e1002888
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W2092692908
Cited by
170
References
63
Citations by year
2026: 6 · 2025: 3 · 2024: 6 · 2023: 11 · 2022: 13 · 2021: 11 · 2020: 18 · 2019: 21 · 2018: 19 · 2017: 11 · 2016: 10 · 2015: 15 · 2014: 20 · 2013: 6
FWCI
7.61
Type
journal article
Language
en
Open access
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • MPG.PuRe (Max Planck Society) (submittedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Max Planck Digital Library (submittedVersion, cc-by)
  • Figshare (submittedVersion, cc-by)
  • Minerva Access (University of Melbourne) (submittedVersion, cc-by)
  • Spiral (Imperial College London) (submittedVersion, other-oa)
Funders
  • Wellcome Trust
  • Medical Research Council
  • Biotechnology and Biological Sciences Research Council (BB/G001863/1, BB/G530268/1, BB/G020434/1, BB/G007934/1)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Physical Sciences › Engineering › Control and Systems Engineering › Control Systems and Identification
  • Social Sciences › Decision Sciences › Management Science and Operations Research › Optimal Experimental Design Methods
Keywords
experimental design, information content, Bayesian inference, information theory, parameter estimation, parameter robustness, biological systems, model inference, prior information
MeSH terms
Bayes Theorem, Models, Theoretical, Uncertainty, Systems Biology

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Maximizing the Information Content of Experiments in Systems Biology. https://app.cochat.ai/papers/10.1371/journal.pcbi.1002888
The paper
Liepe, J., Filippi, S., Komorowski, M., & Stumpf, M. P. H. (2013). Maximizing the Information Content of Experiments in Systems Biology. PLoS Computational Biology. https://doi.org/10.1371/journal.pcbi.1002888

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.